Titus Winters 2018-11-10 04:17:10 UTC
Discussed in San Diego: http://wiki.edg.com/bin/view/Wg21sandiego2018/P1272

This goes into <bit>.
SF F N A SA
2 6 3 0 0

Require that bit-width is a power-of-two, forward to LWG for C++20.
SF F N A SA
3 8 0 0 0

I'm going to schedule this in LWG.

@mclow, this might need some CWG attention (at your discretion), similar to bit_cast. For example, what about padding bits in the object representation of an integer?

Interesting idea. Do we have any integral types that do not "completely fill" an even number of bytes? (bool and char/unsigned char, yes, I think we can just call disable them)

@mclow, note that "byte" equals (the size of) char/unsigned char in the C++ standard by definition (and might not be related to an octet, i.e. an 8-bit quantity); see [intro.memory].
For example, an implementation with 8-bit chars and 31-bit integers would be conforming. With the byte-swap feature, you'd presumably read one padding bit (unspecified value) and turn it into a value bit. Should we prevent that?
My view: Yes, let's make sure the facility does what it should do on today's architectures, and let's just remove the funny stuff from the overload set so that you get a controlled error on alien platforms.

SG6: Leave the issue to other committees.
No objection to unanimous consent.

LWG reviewed this on Thursday AM. U/consent for C++20 (after minor edits

P1272R1 Byteswapping for fun&&nuf (Isabella Muerte)

D1272R3 approved by CWG in Cologne 2019.

Scheduled for LWG review 2021-01-29.

Discussed in LWG 2021-01-29. Author will clarify with core details w.r.t. to padding bits.

https://wiki.edg.com/bin/view/Wg21telecons2021/P1272-20210129

Note to self: CWG are tracking this so LWG should inform them when we decide on a final disposition.

@jwakely The author reach out to me indicating she has no time to work on her papers at the moment.
You remember what core's input was needed for?

@cor3ntin I think @jwakely was setting a reminder, but if you look at the LWG notes from the session there's concerns about the mandates.

Mandates: value does not have padding bits.

The concern was that this is unimplementable by the library. The parameter is passed by value, we have no idea if it came from a bit-field.

However, reading it again, I'm not sure that's a problem. If a bit-field gets promoted and/or converted to an integral type then there are no padding bits, the "extra" bits are zero bits that are part of the value representation.

I think this is forbidding byte swapping something like __int20 which has 20 bits in the value representation and 12 padding bits. The library can enforce that.

I'll check with Core so Izzy doesn't need to.

New revision -- planned as backup paper for re-review on 2021-09-24

https://wiki.edg.com/pub/Wg21telecons2021/LibraryWorkingGroup/D1272R4.html

LWG reviewed and approved during 2021-09-24 telecom

https://wiki.edg.com/bin/view/Wg21telecons2021/P1272-20210924

poll: put P1272r4 into C++23

P1272R4 Byteswapping for fun&&nuf (Isabella Muerte, Corentin Jabot)

Adopted 2021-10.